Although I agree all MT should be switched off in new installations, I doubt MyMemory caches people's source segment text, unless you login using your MyMemory username and select Public TM, or something like that.
This is an important point to be addressed.
MyMemory IS a PUBLIC TM (Public, but company owned), so both MT and public TM results are provided.
The MyMemory API has different parameters.
Searches MyMemory for matches against a segment.
My guess is that CafeTran only uses this parameter (along with Keygen?).
Contributes a translation unit (segment and translation) in some language pair to MyMemory.
Without specifying any key parameter, the contribution is available for everybody.
If a key parameter is specified, the translation will be submitted into your private TM only.
For distinction between Public Data and Private Data, as well as other limitations and terms, see also:
Usage limits: https://mymemory.translated.net/doc/usagelimits.php
So this boils down to CT's implementation.
Igor should publicly confirm SET parameter is not used in CT's API implementation, and MyMemory is only used to search MyMemory archives (and use its Google Translate/Microsoft translator results).
Most novices won't dive as deep as this into this topic. They could be irritated and annoyed by the automatic switched on My Memory function.
Although it is a nice use case, it should be switched off.
If there was a (skippable) tutorial or introduction for the first start of the program to set the first parameters, this would be a nice example to let the user decide. The newer memoQ 2015 versions are shipped with this kind of introduction.
> My guess is that CafeTran only uses this parameter (along with Keygen?).
Yes, CafeTran connects to MyMemory only via GET method (searching only).